Quality check procedure
|last update: Nov. 2015|
Prior to the comparison of a dataset to reference data (2)(5)(6), you need to be sure that these reference data are of sufficiently good quality to serve as reference.
This page aims at giving a large insight of the procedure adopted by MINES ParisTech and Transvalor in term of quality check of ground station measurements. In particular, the first section is dedicated to the quality control steps applied onto the 1 minute Baseline Surface Radiation Network station data (3)(4).
The quality check procedure is derived from the one published in the deliverable 3.2 on data harmonization provided within the framework of the European FP7 ENDORSE project. The table available page 60, also published in the poster(1), summarizes the quality check procedure for all the meteorological components. All the differences between our implementation and this document are highlighted in orange.
NB: This quality check procedure described below is for sub-hourly measurements, which is typically the procedure adopted onto the 1 min BSRN station data. If you need to handle ground measurements with other time steps (such as hourly, or daily values), you need to switch to the adequate column in the above p60 table.
Let's define GHIGRD, DHIGRD and BNIGRD respectively the sub-hourly GHI, DHI and BNI data for the ground station. SZA is the Solar Zénithal angle and SEA the Solar Elevation angle (please note that SEA=90°-SZA). Let's name GHITOA and BNITOA respectively the Global and the Direct component at the Top of Atmosphere.
The below temporal aggregation procedure corresponds to the comparison of a HelioClim-3 (HC3) dataset against 1 min ground station measurements (typically BSRN).
Discussion: If the time step of your ground station measurements or of your estimation is different, this procedure needs to be adapted.
For instance, if we collected hourly in-situ measurements to carry out a comparison with HC3, we must generate hourly HC3 values from the 15 min HC3 values using an intelligent interpolation like in step 2.
Another example: the first HC3 Similarity forecast datasets were generated with a hourly time step. We had to generate the hourly BSRN values from the 1 min data by using once more the step 2, if at least 85% of the 1 min slots are available in the hour.
Then follow the previous steps from step 7 to the end.
|I_ "What are 'partial sums'? Why did you opt for partial sums instead of averages?"|
If you apply any kind of averaging on measurements, you introduce an information which doesn't come from a real measurement. That is why we decided to avoid the distorsion of the values by summing up only the available values. This leads to partial sums for the day or the month. This means also that the resulting value is not representive of an actual daily or monthly value.
|II_ "How did you choose the thresholds in your temporal aggregation procedure?"|
Originally, we didn't set any threshold, and created aggregated values even if only one slot was available. Even if partial sums are not representative of, for instance, a daily or a monthly value, we decided that at least 75% of the 15 min slots were necessary to generate the hourly value, and 75% of the 15 min data to generate the daily or monthly value.
NB: discussion for the hour from 15 min slots: when the sun is rising, you could have less than 4 valid 15 min slots to generate the hourly value. As a consequence, the 75% threshold accepts 1 available slot over 1 valid slot, 2 over 2, 3 over 3 and 4 over 4, plus 3 over 4 valid slots. The first discarded is 2 available over 3 valid slots.
|III_ "Why do you sum up 15 min slots to generate the daily and monthly values?"|
After several tests trying to sum up the hourly values to generate the daily value, we realized that two many days were discarded. Summing up the 15 min slots returns more valid days for the comparisons.
|IV_ "Why did you discard all the GHIGRD values below 50 Wm-2?"|
This decision was motivated by the fact that most instruments present a high uncertainty in low radiation conditions. Moreover, as the consistency checks are not applied for values below this threshold, we finally decided to discard the values.
(In alphabetic order)
(1) Espinar B., P. Blanc, L. Wald, C. Hoyer-Klick, M. Schroedter-Homscheidt, T. Wanderer, 2012. "Controlling the quality of measurements of meteorological variables and solar radiation. From sub-hourly to monthly average time periods", European Geosciences Union 2012, Vienna (Austria) 22–27 April 2012. Poster
(2) ISO Guide to the Expression of Uncertainty in Measurement: first edition, International Organization for Standardization, Geneva, Switzerland, 1995.
(3) Ohmura A., E. G. Dutton, B. Forgan, C. Fröhlich, H. Gilgen, H. Hegner, A. Heimo, G. König-Langlo, B. McArthur, G. Müller, R. Philipona, R. Pinker, C. H. Whitlock, K. Dehne, and . Wild, 1998. "Baseline Surface Radiation Network (BSRN/WCRP): New Precision Radiometry for Climate Research", Bulletin of the American Meteorological Society, vol. 79, N°10, pp. 2115-2136. Article
Content: Table 3 p 2119 gives the uncertainty on the GHI, DHI and BHI components measured by the BSRN stations: +/-5 W/m² for the GHI and the DHI components, and +/-2 W/m² for the BHI. This is why a threshold of respectively 10 W/m² and 4 W/m² is applied on the measurements in the adopted quality check procedure.
(4) Roesch, A., Wild, M., Ohmura, A., Dutton, E.G., Long, C.N., and Zhang, T.: Assessment of BSRN radiation records for the computation of monthly means, Atmos. Meas. Tech., 4, 339–354, doi:10.5194/amt-4-339-2011, 2011. Corrigendum, Atmos. Meas. Tech., 4, 973-973, doi:10.5194/amt-4-973-2011, 2011.
(5) WMO, 1981. Technical Note N° 172, WMO-No. 554, World Meteorological Organization, Geneva, Switzerland, pp. 121-123.
(6) WMO: Guide to meteorological instruments and methods of observation, World Meteorological Organization, WMO - No 8, 7th Edn., Geneva, Switzerland, 2008.